Run length coding accelerator and method for sparse CNN neural network model

Abstract

The invention discloses a run length coding accelerator for a sparse CNN neural network model and a method thereof. The accelerator comprises a top layer controller which is used for identifying the type of input and output data, carrying out grouping processing, and transmitting weight data and excitation data to a run length coding module; a run length encoding module which is used for compressing result data output by the calculation module and transmitting a calculation result after compression encoding to an external memory; a run decoding module which is used for decompressing the data read from the external memory and transmitting the data to the data gating; a data gating which is used for identifying zero values of the input excitation data and the weight data and skipping a multiplication and addition operation for the zero values; and a calculation module which is used for executing multiplication and addition operation on the weight data and the excitation data transmitted by the data gating and giving a calculation result. Storage space and power consumption required by the CNN model are reduced on the basis of data compression and data gating, so that the method has the advantages of high energy efficiency, no need of accessing a large amount of memory and the like.

Description

technical field [0001] The invention relates to a run-length coding accelerator and a method thereof for a sparse CNN neural network model, and belongs to the field of digital signal processing and hardware accelerated neural network algorithms. Background technique [0002] Neural network is currently a research hotspot in the field of artificial intelligence. LeNet is one of the earliest convolutional neural networks and has promoted the development of deep learning. Due to the influence of computer performance, although LeNet has achieved good results in image classification, it has not attracted attention. CNN deepens the network structure on the basis of LeNet and learns richer and higher-dimensional image features. CNN uses a deeper network structure, namely convolutional layer + convolutional layer + pooling layer to extract the features of the image, and replaces the previous sigmoid with Relu as the activation function. On ImageNet LSVRC-2010 in 2010, CNN achieved...

AI Technical Summary

Problems solved by technology

[0006] In order to overcome the problem of a large amount of memory access and a large amount of energy consumption in the process of calculating the neural network in the memory, the present invention proposes an accelerator using run-length coding and its acceleration method for the sparse CNN model

Images